ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1
3075 commits 1 branch 0 tags 37 MiB
Commit graph

776 commits

Author SHA1 Message Date
Xin Qiu
1dd40b429c
enable fp4 fused mlp and qkv (#10531) 
* enable fp4 fused mlp and qkv

* update qwen

* update qwen2
 2024-03-26 08:34:00 +08:00
Wang, Jian4
16b2ef49c6
Update_document by heyang (#30) 2024-03-25 10:06:02 +08:00
Wang, Jian4
a1048ca7f6
Update setup.py and add new actions and add compatible mode (#25) 
* update setup.py

* add new action

* add compatible mode
 2024-03-22 15:44:59 +08:00
Wang, Jian4
9df70d95eb
Refactor bigdl.llm to ipex_llm (#24) 
* Rename bigdl/llm to ipex_llm

* rm python/llm/src/bigdl

* from bigdl.llm to from ipex_llm
 2024-03-22 15:41:21 +08:00
Wang, Jian4
34d0a9328c LLM: Speed-up mixtral in pipeline parallel inference (#10472) 
* speed-up mixtral

* fix style
 2024-03-22 11:06:28 +08:00
Cengguang Zhang
b9d4280892 LLM: fix baichuan7b quantize kv abnormal output. (#10504) 
* fix abnormal output.

* fix style.

* fix style.
 2024-03-22 10:00:08 +08:00
Yishuo Wang
f0f317b6cf fix a typo in yuan (#10503) 2024-03-22 09:40:04 +08:00
Guancheng Fu
3a3756b51d Add FastChat bigdl_worker (#10493) 
* done

* fix format

* add licence

* done

* fix doc

* refactor folder

* add license
 2024-03-21 18:35:05 +08:00
Xin Qiu
dba7ddaab3 add sdp fp8 for qwen llama436 baichuan mistral baichuan2 (#10485) 
* add sdp fp8

* fix style

* fix qwen

* fix baichuan 13

* revert baichuan 13b and baichuan2-13b

* fix style

* update
 2024-03-21 17:23:05 +08:00
Kai Huang
30f111cd32 lm_head empty_cache for more models (#10490) 
* modify constraint

* fix style
 2024-03-21 17:11:43 +08:00
binbin Deng
2958ca49c0 LLM: add patching function for llm finetuning (#10247) 2024-03-21 16:01:01 +08:00
Kai Huang
021d77fd22 Remove softmax upcast fp32 in llama (#10481) 
* update

* fix style
 2024-03-20 18:17:34 +08:00
Yishuo Wang
cfdf8ad496 Fix modules_not_to_convert argument (#10483) 2024-03-20 17:47:03 +08:00
Xiangyu Tian
cbe24cc7e6 LLM: Enable BigDL IPEX Int8 (#10480) 
Enable BigDL IPEX Int8
 2024-03-20 15:59:54 +08:00
ZehuaCao
1d062e24db Update serving doc (#10475) 
* update serving doc

* add tob

* update

* update

* update

* update vllm worker
 2024-03-20 14:44:43 +08:00
Cengguang Zhang
4581e4f17f LLM: fix whiper model missing config. (#10473) 
* fix whiper model missing config.

* fix style.

* fix style.

* style.
 2024-03-20 14:22:37 +08:00
Yishuo Wang
749bedaf1e fix rwkv v5 fp16 (#10474) 2024-03-20 13:15:08 +08:00
Yuwen Hu
72bcc27da9 [LLM] Add TransformersBgeEmbeddings class in bigdl.llm.langchain.embeddings (#10459) 
* Add TransformersBgeEmbeddings class in bigdl.llm.langchain.embeddings

* Small fixes
 2024-03-19 18:04:35 +08:00
Cengguang Zhang
463a86cd5d LLM: fix qwen-vl interpolation gpu abnormal results. (#10457) 
* fix qwen-vl interpolation gpu abnormal results.

* fix style.

* update qwen-vl gpu example.

* fix comment and update example.

* fix style.
 2024-03-19 16:59:39 +08:00
Xin Qiu
bbd749dceb qwen2 fp8 cache (#10446) 
* qwen2 fp8 cache

* fix style check
 2024-03-19 08:32:39 +08:00
Yang Wang
9e763b049c Support running pipeline parallel inference by vertically partitioning model to different devices (#10392) 
* support pipeline parallel inference

* fix logging

* remove benchmark file

* fic

* need to warmup twice

* support qwen and qwen2

* fix lint

* remove genxir

* refine
 2024-03-18 13:04:45 -07:00
Xiangyu Tian
dbdeaddd6a LLM: Fix log condition for BIGDL_OPT_IPEX (#10441) 
remove log for BIGDL_OPT_IPEX
 2024-03-18 16:03:51 +08:00
Xin Qiu
399843faf0 Baichuan 7b fp16 sdp and qwen2 pvc sdp (#10435) 
* add baichuan sdp

* update

* baichuan2

* fix

* fix style

* revert 13b

* revert
 2024-03-18 10:15:34 +08:00
Yishuo Wang
bd64488b2a add mask support for llama/chatglm fp8 sdp (#10433) 
* add mask support for fp8 sdp

* fix chatglm2 dtype

* update
 2024-03-15 17:36:52 +08:00
Xin Qiu
24473e331a Qwen2 fp16 sdp (#10427) 
* qwen2 sdp and refine

* update

* update

* fix style

* remove use_flash_attention
 2024-03-15 13:12:03 +08:00
Ruonan Wang
b036205be2 LLM: add fp8 sdp for chatglm2/3 (#10411) 
* add fp8 sdp for chatglm2

* fix style
 2024-03-15 09:38:18 +08:00
Wang, Jian4
fe8976a00f LLM: Support gguf models use low_bit and fix no json(#10408) 
* support others model use low_bit

* update readme

* update to add *.json
 2024-03-15 09:34:18 +08:00
Xin Qiu
cda38f85a9 Qwen fp16 sdp (#10401) 
* qwen sdp

* fix

* update

* update

* update sdp

* update

* fix style check

* add to origin type
 2024-03-15 08:51:50 +08:00
dingbaorong
1c0f7ed3fa add xpu support (#10419) 2024-03-14 17:13:48 +08:00
Heyang Sun
7d29765092 refactor qwen2 forward to enable XPU (#10409) 
* refactor awen2 forward to enable XPU

* Update qwen2.py
 2024-03-14 11:03:05 +08:00
ZehuaCao
f66329e35d Fix multiple get_enable_ipex function error (#10400) 
* fix multiple get_enable_ipex function error

* remove get_enable_ipex_low_bit function
 2024-03-14 10:14:13 +08:00
Kai Huang
76e30d8ec8 Empty cache for lm_head (#10317) 
* empty cache

* add comments
 2024-03-13 20:31:53 +08:00
Yishuo Wang
06a851afa9 support new baichuan model (#10404) 2024-03-13 17:45:50 +08:00
Yishuo Wang
b268baafd6 use fp8 sdp in llama (#10396) 2024-03-13 16:45:38 +08:00
Xiangyu Tian
60043a3ae8 LLM: Support Baichuan2-13b in BigDL-vLLM (#10398) 
Support Baichuan2-13b in BigDL-vLLM.
 2024-03-13 16:21:06 +08:00
Xiangyu Tian
e10de2c42d [Fix] LLM: Fix condition check error for speculative decoding on CPU (#10402) 
Fix condition check error for speculative decoding on CPU
 2024-03-13 16:05:06 +08:00
Heyang Sun
d72c0fad0d Qwen2 SDPA forward on CPU (#10395) 
* Fix Qwen1.5 CPU forward

* Update convert.py

* Update qwen2.py
 2024-03-13 13:10:03 +08:00
Wang, Jian4
0193f29411 LLM : Enable gguf float16 and Yuan2 model (#10372) 
* enable float16

* add yun files

* enable yun

* enable set low_bit on yuan2

* update

* update license

* update generate

* update readme

* update python style

* update
 2024-03-13 10:19:18 +08:00
Yina Chen
f5d65203c0 First token lm_head optimization (#10318) 
* add lm head linear

* update

* address comments and fix style

* address comment
 2024-03-13 10:11:32 +08:00
Xin Qiu
28c4a8cf5c Qwen fused qkv (#10368) 
* fused qkv + rope for qwen

* quantized kv cache

* fix

* update qwen

* fixed quantized qkv

* fix

* meet code review

* update split

* convert.py

* extend when no enough kv

* fix
 2024-03-12 17:39:00 +08:00
Yishuo Wang
741c2bf1df use new rms norm (#10384) 2024-03-12 17:29:51 +08:00
Xiangyu Tian
0ded0b4b13 LLM: Enable BigDL IPEX optimization for int4 (#10319) 
Enable BigDL IPEX optimization for int4
 2024-03-12 17:08:50 +08:00
Zhao Changmin
df2b84f7de Enable kv cache on arc batch (#10308) 2024-03-12 16:46:04 +08:00
Guancheng Fu
cc4148636d [FastChat-integration] Add initial implementation for loader (#10323) 
* add initial implementation for loader

* add test method for model_loader

* data

* Refine
 2024-03-12 10:54:59 +08:00
binbin Deng
dbcfc5c2fa LLM: fix error of 'AI-ModelScope/phi-2' hosted by ModelScope hub (#10364) 2024-03-11 16:19:17 +08:00
Chen, Zhentao
a425eaabfc fix from_pretrained when device_map=None (#10361) 
* pr trigger

* fix error when device_map=None

* fix device_map=None
 2024-03-11 16:06:12 +08:00
Yina Chen
d7b765fd3f serving xpu memory opt (#10358) 2024-03-11 15:21:22 +08:00
Ruonan Wang
be29833b2b LLM: fix qwen2 (#10356) 2024-03-11 09:29:08 +08:00
Zhicun
9026c08633 Fix llamaindex AutoTokenizer bug (#10345) 
* fix tokenizer

* fix AutoTokenizer bug

* modify code style
 2024-03-08 16:24:50 +08:00
Keyan (Kyrie) Zhang
7a621a4db0 Fix device_map bug by raise an error when using device_map=xpu (#10340) 
* Fix device_map bug by raise an error when using device_map=xpu

* Fix sync error

* Fix python style

* Use invalidInputError instead of invalidOperationError
 2024-03-08 13:38:52 +08:00
Yishuo Wang
1ac193ba02 add rope theta argument (#10343) 2024-03-07 17:27:19 +08:00
Cengguang Zhang
496d18ab6d LLM: add quantize kv cache support for baichuan 7b and 13b. (#10330) 
* add quantize kv cache for baichuan 7b and 13b.

* fix typo.

* fix.

* fix style.

* fix style.
 2024-03-07 16:17:38 +08:00
Yina Chen
9ea499ca68 Optimize speculative decoding PVC memory usage (#10329) 
* optimize memory

* update

* update

* update

* support other models

* update

* fix style
 2024-03-06 09:54:21 +08:00
dingbaorong
cc796848ea fix typos (#10274) 
Co-authored-by: Ariadne <wyn2000330@126.com>
 2024-03-05 18:38:22 +08:00
Yishuo Wang
0011ff9f64 optimize bge large performance (#10324) 2024-03-05 17:06:03 +08:00
Cengguang Zhang
30d009bca7 LLM: support quantized kv cache for Mistral in transformers >=4.36.0 (#10326) 
* support quantize kv for mistral in transformers 4.36

* update mistral support.

* fix style.
 2024-03-05 16:23:50 +08:00
dingbaorong
1e6f0c6f1a Add llamaindex gpu example (#10314) 
* add llamaindex example

* fix core dump

* refine readme

* add trouble shooting

* refine readme

---------

Co-authored-by: Ariadne <wyn2000330@126.com>
 2024-03-05 13:36:00 +08:00
dingbaorong
fc7f10cd12 add langchain gpu example (#10277) 
* first draft

* fix

* add readme for transformer_int4_gpu

* fix doc

* check device_map

* add arc ut test

* fix ut test

* fix langchain ut

* Refine README

* fix gpu mem too high

* fix ut test

---------

Co-authored-by: Ariadne <wyn2000330@126.com>
 2024-03-05 13:33:57 +08:00
Cengguang Zhang
ab9fc2485f LLM: add quantize kv support for llama transformer 4.36 (#10298) 
* add quantize kv support for llama transformer 4.36

* fix style.

* fix style.
 2024-03-04 10:33:35 +08:00
SONG Ge
0ab40917fb [LLM] Split merged_qk to separated q/k linear (#10299) 
* modify merge_qk_linear to separated q/k linear

* update
 2024-03-01 16:48:55 +08:00
Yang Wang
f4d7dbcde2 use fused qkv forward in qwen2 (#10185) 
* use fused qkv forward in qwen2

* support both

* fix style

* fix rope

* remove pring

* fix style

* clean up
 2024-03-01 16:46:35 +08:00
Wang, Jian4
beb9433cec LLM: Reduce speculative _ipex_optimize_model memory use (#10281) 
* use tpp

* update ipex
 2024-03-01 13:48:23 +08:00
Yuwen Hu
f0ff0eebe1 [LLM] Support quantize kv cache for Baichuan2 7B (#10280) 
* Add quatized kv cache framework for Baichuan2 7B

* Support quantize kv cache for baichuan2

* Small fix

* Fix python style
 2024-03-01 13:35:42 +08:00
SONG Ge
273de341d7 hot-fix silu error import (#10292) 2024-03-01 10:11:37 +08:00
Xin Qiu
232273a1b5 Enable Gemma fused mlp + Gelu (#10276) 
* update llama mlp forward

* add all

* fix style check

* split

* update

* update

* update

* fix style
 2024-02-29 16:53:24 +08:00
Guancheng Fu
2d930bdca8 Add vLLM bf16 support (#10278) 
* add argument load_in_low_bit

* add docs

* modify gpu doc

* done

---------

Co-authored-by: ivy-lv11 <lvzc@lamda.nju.edu.cn>
 2024-02-29 16:33:42 +08:00
SONG Ge
13b0bc9075 [LLM] Add quantize_kv optimization for yuan2 model (#10243) 
* add initial quantize_kv support for yuan2 model

* fix yuan2 quantize_kv generation

* apply fp16 conv layer optimizations

* disable mlp for quantize_kv
 2024-02-29 16:33:26 +08:00
Zhicun
4e6cc424f1 Add LlamaIndex RAG (#10263) 
* run demo

* format code

* add llamaindex

* add custom LLM with bigdl

* update

* add readme

* begin ut

* add unit test

* add license

* add license

* revised

* update

* modify docs

* remove data folder

* update

* modify prompt

* fixed

* fixed

* fixed
 2024-02-29 15:21:19 +08:00
Ruonan Wang
a9fd20b6ba LLM: Update qkv fusion for GGUF-IQ2 (#10271) 
* first commit

* update mistral

* fix transformers==4.36.0

* fix

* disable qk for mixtral now

* fix style
 2024-02-29 12:49:53 +08:00
Ruonan Wang
4b08bc1417 LLM: relax batch check of flash atttention by double check attention mask (#10270) 
* relax batch check

* fix

* fix style
 2024-02-29 09:39:55 +08:00
Yina Chen
07f36fbfcc Fix gptj failed to extend (#10269) 2024-02-29 09:39:27 +08:00
Yishuo Wang
cccb02dad1 fix baichuan2 13b 2k input (#10267) 2024-02-28 17:20:20 +08:00
Heyang Sun
7244fd1ba5 Fix Arc StarCoder wrong query_shape when input is long (#10268) 
* Fix Arc StarCoder wrong query_shape when input is long

* Update gptbigcode.py
 2024-02-28 17:07:08 +08:00
Cengguang Zhang
a4de3095f3 LLM: Support quantize kv cache in mistral. (#10261) 
* init

* update quantize kv.
 2024-02-28 14:08:08 +08:00
Zhicun
308e637d0d Add DeepSeek-MoE-16B-Chat (#10155) 
* dsmoe-hf add

* add dsmoe pytorch

* update README

* modify comment

* remove GPU example

* update model name

* format code
 2024-02-28 10:12:09 +08:00
Yang Wang
c581c6db30 draft mmint4 (#10031) 
change to llm.cpp

support transposed format

revert

implement qkv fuse

fix style

change to vertically pack

change to enable_xetla

fix mlp_fusion_check

remove comments

address comments

add some comments

fix style
 2024-02-27 14:55:16 -08:00
Yishuo Wang
b4fa4ab46f optimize yuan 2.0 again (#10252) 2024-02-27 14:51:42 +08:00
Heyang Sun
36a9e88104 Speculative Starcoder on CPU (#10138) 
* Speculative Starcoder on CPU

* enable kv-cache pre-allocation

* refine codes

* refine

* fix style

* fix style

* fix style

* refine

* refine

* Update speculative.py

* Update gptbigcode.py

* fix style

* Update speculative.py

* enable mixed-datatype layernorm on top of torch API

* adaptive dtype

* Update README.md
 2024-02-27 09:57:29 +08:00
Yishuo Wang
a47989c860 optimize yuan 2.0 performance (#10244) 2024-02-26 17:20:10 +08:00
Wang, Jian4
6c74b99a28 LLM: Update qwen readme (#10245) 2024-02-26 17:03:09 +08:00
Wang, Jian4
f9b75f900b LLM: Enable qwen target_model ipex (#10232) 
* change order

* enable qwen ipex

* update qwen example

* update

* fix style

* update
 2024-02-26 16:41:12 +08:00
Yuwen Hu
e38e29511c [LLM] Yuan2 MLP and Rotary optimization (#10231) 
* Add optimization for rotary embedding

* Add mlp fused optimizatgion

* Python style fix

* Fix rotary embedding due to logits difference

* Small fix
 2024-02-26 15:10:08 +08:00
SONG Ge
df2f3885ba [LLM] Enable kv_cache and forward_qkv optimizations for yuan2 (#10225) 
* add init kv_cache support for yuan2

* add forward qkv in yuan
 2024-02-26 11:29:48 +08:00
Ruonan Wang
28513f3978 LLM: support fp16 embedding & add mlp fusion for iq2_xxs (#10219) 
* add fp16 embed

* small fixes

* fix style

* fix style

* fix comment
 2024-02-23 17:26:24 +08:00
Yuwen Hu
eeecd9fc08 Python style fix (#10230) 2024-02-23 17:21:23 +08:00
Yuwen Hu
e511bbd8f1 [LLM] Add basic optimization framework for Yuan2 (#10227) 
* Add basic optimization framework for Yuan2

* Small fix

* Python style fix

* Small fix

* Small fix
 2024-02-23 17:05:00 +08:00
Xin Qiu
30795bdfbc Gemma optimization: rms_norm, kv_cache, fused_rope, fused_rope+qkv (#10212) 
* gemma optimization

* update

* update

* fix style

* meet code review
 2024-02-23 10:07:24 +08:00
Guoqiong Song
63681af97e falcon for transformers 4.36 (#9960) 
* falcon for transformers 4.36
 2024-02-22 17:04:40 -08:00
Yina Chen
ce5840a8b7 GPT-J rope optimization on xpu (#10182) 
* optimize

* update

* fix style & move use_fuse_rope

* add ipex version check

* fix style

* update

* fix style

* meet comments

* address comments

* fix style
 2024-02-22 16:25:12 +08:00
Xiangyu Tian
f445217d02 LLM: Update IPEX to 2.2.0+cpu and Refactor for _ipex_optimize (#10189) 
Update IPEX to 2.2.0+cpu and refactor for _ipex_optimize.
 2024-02-22 16:01:11 +08:00
Heyang Sun
c876d9b5ca Support for MPT rotary embedding (#10208) 2024-02-22 15:16:31 +08:00
Ruonan Wang
5e1fee5e05 LLM: add GGUF-IQ2 examples (#10207) 
* add iq2 examples

* small fix

* meet code review

* fix

* meet review

* small fix
 2024-02-22 14:18:45 +08:00
SONG Ge
ca1166a0e5 [LLM] Add quantize kv_cache for Baichuan2-13B (#10203) 
* add quantize kv_cache for baichuan2-13b

* style fix
 2024-02-22 13:43:35 +08:00
Ruonan Wang
34ee1aa91f LLM: add esimd sdp support for chatglm3 (#10205) 
* add esimd sdp support

* fix style
 2024-02-22 13:37:16 +08:00
Ruonan Wang
f7c96b19ef LLM: support iq2 for mixtral (#10191) 
* support name mapping for mixtral

* support mixtral mixed quantization

* fix style

* fix
 2024-02-21 16:00:29 +08:00
Xin Qiu
56ad781f2f qwen2 cpu fix (#10187) 2024-02-21 11:23:51 +08:00
Zhao Changmin
4fbf449c2d for rwkv4 (#10179) 2024-02-21 10:11:10 +08:00
Ruonan Wang
3288acb8de LLM : Support embedding quantization (only q2k now) (#10170) 
* basic logic added

* basic support

* support save&load, update mixed strategy

* fix style

* use int8 for lm_head

* add check for xpu
 2024-02-20 16:56:57 +08:00
binbin Deng
2bb96c775c LLM: fix device setting during saving optimized model (#10154) 2024-02-20 09:52:59 +08:00
Xin Qiu
1f6d5b9f30 enable fused rmsnorm and rope qwen2 (#10163) 
* qwen2

* change convert

* cleanup
 2024-02-20 08:33:09 +08:00
Zhao Changmin
f8730e8dc1 Skip rescale rwkv linear when load_low_bit (#10164) 
* rwkv_ld
 2024-02-19 15:56:42 +08:00
Heyang Sun
3e2af5ec0a Fix IPEX Baichuan Speculative (#10162) 
* Fix IPEX Baichuan Speculative

* compatible with 13B

* Update speculative.py
 2024-02-19 15:27:34 +08:00
Yina Chen
23c91cdce6 [LLM] Add min_step_draft in speculative decoding (#10142) 
* Fix gptj kvcache & position id

* Add min_draft_tokens in speculative decoding

* fix style

* update
 2024-02-19 14:31:41 +08:00
Wang, Jian4
f2417e083c LLM: enable chatglm3-6b target_model ipex (#10085) 
* init

* always make casual_mask

* not return last tensor

* update

* optimize_model = False

* enable optimized=False

* enable optimized_model=true

* speed_up ipex target_model

* remove if True

* use group_size

* update python style

* update

* update
 2024-02-19 13:38:32 +08:00
Yina Chen
1508d6b089 Fix gptj kvcache & position id (#10141) 2024-02-18 10:02:49 +08:00
Yishuo Wang
4d33aac7f9 quick fix qwen2 fp8 kv cache (#10135) 2024-02-08 17:04:59 +08:00
Cengguang Zhang
39d90839aa LLM: add quantize kv cache for llama. (#10086) 
* feat: add quantize kv cache for llama.

* fix style.

* add quantized attention forward function.

* revert style.

* fix style.

* fix style.

* update quantized kv cache and add quantize_qkv

* fix style.

* fix style.

* optimize quantize kv cache.

* fix style.
 2024-02-08 16:49:22 +08:00
Yishuo Wang
d848efe17c add quantize kv cache support for qwen2 (#10134) 2024-02-08 16:17:21 +08:00
SONG Ge
3f79128ed7 [LLM] Enable kv_cache optimization for Qwen2 on transformers-v4.37.0 (#10131) 
* add support for kv_cache optimization on transformers-v4.37.0

* enable attention forward

* style fix

* disable rotary for now
 2024-02-08 14:20:26 +08:00
Ruonan Wang
063dc145ac LLM: basic support for q2k (#10132) 
* basic support for q2k

* fix style
 2024-02-08 13:52:01 +08:00
Cengguang Zhang
0cf6a12691 LLM: add default torch_dtype for fp16. (#10124) 
* set default torch_dtype for fp16.

* fix style.

* bug fix.

* update bug fix.
 2024-02-08 10:24:16 +08:00
Yishuo Wang
1aa0c623ce disable fused layer norm on UHD (#10130) 2024-02-08 10:20:01 +08:00
Yuwen Hu
a8450fc300 [LLM] Support MLP optimization for Qwen1.5 (#10123) 2024-02-08 09:15:34 +08:00
binbin Deng
925f82107e LLM: support models hosted by modelscope (#10106) 2024-02-07 16:46:36 +08:00
Xiangyu Tian
8953acd7d6 [LLM] Fix log condition for BIGDL_OPT_IPEX (#10115) 
Fix log condition for BIGDL_OPT_IPEX
 2024-02-07 10:27:10 +08:00
Yuwen Hu
518ef95abc Small fix for Nonetype error (#10104) 2024-02-06 14:58:52 +08:00
Ruonan Wang
d61f4905ac LLM: 2bit quantization initial support (#10042) 
* basis quantize support

* fix new module name

* small update

* and mixed int4 with iq2_xxs

* remove print

* code refactor

* fix style

* meet code review
 2024-02-06 14:58:32 +08:00
Jiao Wang
33b9e7744d fix dimension (#10097) 2024-02-05 15:07:38 -08:00
Zhicun
7d2be7994f add phixtral and optimize phi-moe (#10052) 2024-02-05 11:12:47 +08:00
Zhicun
676d6923f2 LLM: modify transformersembeddings.embed() in langchain (#10051) 2024-02-05 10:42:10 +08:00
Jin Qiao
ad050107b3 LLM: fix mpt load_low_bit issue (#10075) 
* fix

* retry

* retry
 2024-02-05 10:17:07 +08:00
Ruonan Wang
8e33cb0f38 LLM: support speecht5_tts (#10077) 
* support speecht5_tts

* fix
 2024-02-04 13:26:42 +08:00
ivy-lv11
428b7105f6 Add HF and PyTorch example InternLM2 (#10061) 2024-02-04 10:25:55 +08:00
Yina Chen
77be19bb97 LLM: Support gpt-j in speculative decoding (#10067) 
* gptj

* support gptj in speculative decoding

* fix

* update readme

* small fix
 2024-02-02 14:54:55 +08:00
Xin Qiu
6e0f1a1e92 use apply_rotary_pos_emb_cache_freq_xpu in mixtral (#10060) 
* use apply_rotary_pos_emb_cache_freq_xpu in mixtral

* fix style
 2024-02-01 15:40:49 +08:00
Heyang Sun
601024f418 Mistral CPU example of speculative decoding (#10024) 
* Mistral CPU example of speculative decoding

* update transformres version

* update example

* Update README.md
 2024-02-01 10:52:32 +08:00
Heyang Sun
968e70544d Enable IPEX Mistral in Speculative (#10059) 2024-02-01 10:48:16 +08:00
Yina Chen
3ca03d4e97 Add deepmind sample into bigdl-llm speculative decoding (#10041) 
* migrate deepmind sample

* update

* meet comments

* fix style

* fix style
 2024-02-01 09:57:02 +08:00
Wang, Jian4
7e5cd42a5c LLM : Update optimize ipex bf16 (#10038) 
* use 4.35.2 and remove

* update rmsnorm

* remove

* remove

* update python style

* update

* update python style

* update

* fix style

* update

* remove whitespace
 2024-01-31 10:59:55 +08:00
Ruonan Wang
3685622f29 LLM: fix llama 4.36 forward(#10047) 2024-01-31 10:31:10 +08:00
Yishuo Wang
53a5140eff Optimize rwkv v5 rest token again (#10043) 2024-01-31 10:01:11 +08:00
Ruonan Wang
6b63ba23d1 LLM: add full module name during convert (#10035) 2024-01-30 14:43:07 +08:00
Yishuo Wang
7dfa6dbe46 add rwkv time shift optimization (#10032) 2024-01-30 14:10:55 +08:00
Xiangyu Tian
f57d0fda8b [LLM] Use IPEX Optimization for Self Speculative Decoding (#9997) 
Use IPEX Optimization for Self Speculative Decoding
 2024-01-30 09:11:06 +08:00
Ruonan Wang
ccf8f613fb LLM: update fp16 Linear on ARC/FLEX (#10023) 2024-01-29 18:25:26 +08:00
Shaojun Liu
824c8029d7 Fix "local variable 'model' referenced before assignment" (#10022) 2024-01-29 16:18:04 +08:00
Xiangyu Tian
f37e4702bc [LLM] Use IPEX Optimization for BF16 Model (#9988) 
Use IPEX Optimization for BF16 Model by env BIGDL_OPT_IPEX=true
 2024-01-29 11:28:25 +08:00
Yishuo Wang
d720554d43 simplify quantize kv cache api (#10011) 2024-01-29 09:23:57 +08:00
Yina Chen
a3322e2a6c add fp8 e5 to use_xmx (#10015) 2024-01-26 18:29:46 +08:00
Qiyuan Gong
9e18ea187f [LLM] Avoid KV Cache OOM when seq len is larger than 1 (#10006) 
* Avoid OOM during muti-round streaming chat with kv cache
* For llama like kv cache, i.e., [bs, n_head, seq_len, head_dim], use is_enough_kv_cache_room_4_31.
* Other models need to compare kv cache size with kv_len.
 2024-01-26 17:30:08 +08:00
Ruonan Wang
a00efa0564 LLM: add mlp & qkv fusion for FP16 Llama-7B (#9932) 
* add mlp fusion for llama

* add mlp fusion

* fix style

* update

* add mm_qkv_out

* fix style

* update

* meet code review

* meet code review
 2024-01-26 11:50:38 +08:00
Wang, Jian4
98ea3459e5 LLM : Fix llama draft_model dtype error (#10005) 
* fix llama draft_model dtype error

* updat
 2024-01-26 10:59:48 +08:00
Yishuo Wang
aae1870096 fix qwen kv cache length (#9998) 2024-01-26 10:15:01 +08:00
Yishuo Wang
24b34b6e46 change xmx condition (#10000) 2024-01-25 17:48:11 +08:00
Yishuo Wang
bf65548d29 Add quantize kv cache support for chaglm2/3 (#9996) 2024-01-25 16:55:59 +08:00
Wang, Jian4
9bff84e6fd LLM: Convert draft_model kv_cache from bf16 to fp32 (#9964) 
* convert bf16 to fp32

* update

* change when init

* init first and cut off after

* init and exchange

* update python type

* update

* fix bug

* update

* update
 2024-01-25 11:20:27 +08:00
Yina Chen
27338540c3 Fix repetition_penalty not activated issue (#9989) 2024-01-25 10:40:41 +08:00
Yuwen Hu
b27e5a27b9 Remove the check for meta device in _replace_with_low_bit_linear (#9984) 2024-01-24 18:15:39 +08:00
Yina Chen
b176cad75a LLM: Add baichuan2 gpu spec example (#9973) 
* add baichuan2 gpu spec example

* update readme & example

* remove print

* fix typo

* meet comments

* revert

* update
 2024-01-24 16:40:16 +08:00
Chen, Zhentao
e0db44dcb6 fix unexpected keyword argument 'device' (#9982) 
* add device for chatglm3 only

* add comment for this change

* fix style

* fix style

* fix style again..

* finally fixed style
 2024-01-24 13:20:46 +08:00
Yuwen Hu
8d28aa8e2b [LLM] Fix the model.device problem when cpu_embedding=True (#9971) 
* Overwrite the device attribute for CPUPinnedParam

* Expose cpu_embedding=True for Linux users

* Fix python style
 2024-01-23 18:51:11 +08:00
Yishuo Wang
f82782cd3b fix starcoder (#9975) 2024-01-23 17:24:53 +08:00
Yishuo Wang
2c8a9aaf0d fix qwen causal mask when quantize_kv_cache=True (#9968) 2024-01-23 16:34:05 +08:00
Yina Chen
36c665667d Add logits processor & qwen eos stop in speculative decoding (#9963) 
* add logits processor & qwen eos

* fix style

* fix

* fix

* fix style

* fix style

* support transformers 4.31

* fix style

* fix style

---------

Co-authored-by: rnwang04 <ruonan1.wang@intel.com>
 2024-01-23 15:57:28 +08:00
Xin Qiu
da4687c917 fix fp16 (#9970) 2024-01-23 15:53:32 +08:00
Ruonan Wang
27b19106f3 LLM: add readme for speculative decoding gpu examples (#9961) 
* add readme

* add readme

* meet code review
 2024-01-23 12:54:19 +08:00
Chen, Zhentao
39219b7e9a add default device meta when lcmu enabled (#9941) 2024-01-23 11:00:49 +08:00
Xin Qiu
dacf680294 add fused rotary pos emb for qwen (#9956) 
* add fused rotary pos emb for qwen

* update
 2024-01-23 10:37:56 +08:00
Ruonan Wang
7b1d9ad7c0 LLM: limit esimd sdp usage for k_len < 8 (#9959) 
* update

* fix
 2024-01-23 09:28:23 +08:00
Ruonan Wang
3e601f9a5d LLM: Support speculative decoding in bigdl-llm (#9951) 
* first commit

* fix error, add llama example

* hidden print

* update api usage

* change to api v3

* update

* meet code review

* meet code review, fix style

* add reference, fix style

* fix style

* fix first token time
 2024-01-22 19:14:56 +08:00
Heyang Sun
fb91c97fe8 support for Baichuan/Baichuan2 13B Chat running speculative decoding (#9921) 
* support for Baichuan/Baichuan2 13B Chat running speculative decoding

* fix stype
 2024-01-22 09:11:44 +08:00
Xin Qiu
97f0cd8975 optimize Decilm 7b (#9922) 
* optimize deci

* update

* decilm attension forward
 2024-01-19 17:31:13 +08:00
Wang, Jian4
bcaeb05272 Update optimize qwen (#9943) 
* update for n tokens input

* fix dtype

* update
 2024-01-19 16:54:59 +08:00
Ruonan Wang
bf37b3a670 LLM: optimize CPU speculative decoding of chatglm3 (#9928) 
* update

* fix style

* meet code review
 2024-01-19 14:10:22 +08:00
Shaojun Liu
967714bac8 gguf memory optimization for mixtral (#9939) 2024-01-19 11:13:15 +08:00
Lilac09
7032a2ad73 Optimize gguf load memory for mistral (#9923) 
* optimize gguf load for mistral

* fix output of gguf mistral

* reset
 2024-01-19 09:14:39 +08:00
Shaojun Liu
9a46f019d7 gguf memory optimization for baichuan (#9937) 2024-01-19 09:11:02 +08:00
Guancheng Fu
2e1448f08e [Serving] Add vllm_worker to fastchat serving framework (#9934) 
* add worker

* finish

* finish

* add license

* add more comments
 2024-01-18 21:33:36 +08:00
Yishuo Wang
7bbb98abb6 Disable fused layer norm when using XMX to fix mpt UT (#9933) 2024-01-18 16:22:12 +08:00
Wang, Jian4
1fc9dfa265 LLM: Update for Qwen n tokens inputs (#9931) 
* update for n tokens inputs

* update style

* update
 2024-01-18 15:56:29 +08:00
Heyang Sun
5184f400f9 Fix Mixtral GGUF Wrong Output Issue (#9930) 
* Fix Mixtral GGUF Wrong Output Issue

* fix style

* fix style
 2024-01-18 14:11:27 +08:00
Yishuo Wang
453df868c9 add rwkv v5 attention kernel (#9927) 2024-01-18 10:16:29 +08:00
Ruonan Wang
054952f82f LLM: Fix rope of chatglm3 to support speculative decoding on CPU (#9926) 2024-01-18 09:28:10 +08:00
Ziteng Zhang
18cd1f1432 [LLM]Solve the problem of calling bmm operator in BF16Linear (#9924) 
* Solve the problem of calling bmm operator in BF16Linear
 2024-01-17 18:08:35 +08:00
Yina Chen
98b86f83d4 Support fast rope for training (#9745) 
* init

* init

* fix style

* add test and fix

* address comment

* update

* merge upstream main
 2024-01-17 15:51:38 +08:00
Ruonan Wang
427f75000b LLM: fix sdp of chatglm3 (#9917) 
* fix

* fix

* fix
 2024-01-17 13:37:28 +08:00
Yishuo Wang
94767da7cf optimize rwkv v4 first token performance (#9912) 2024-01-17 09:27:41 +08:00
Shaojun Liu
b909c5c9c2 GGUF load memory optimization (#9913) 
* block-wise

* convert linear for module

* revert

* Fix PEP8 checks Error
 2024-01-16 18:54:39 +08:00
Xin Qiu
dee32f7d15 copy fused rms norm's reuslt to avoid <unk> (#9909) 2024-01-16 16:54:08 +08:00
Ruonan Wang
8d7326ae03 LLM: fix chatglm3 sdp to support speculative decoding (#9900) 
* fix chatglm3

* fix

* update

* meet code review

* fix
 2024-01-16 11:29:13 +08:00
Guancheng Fu
9f34da7cdb Update PVC XMX condition (#9901) 
* update pvc xmx condition

* update condition

* update conditon
 2024-01-15 15:42:15 +08:00
Yishuo Wang
6637860ddf change xmx condition (#9896) 2024-01-12 19:51:48 +08:00
Ruonan Wang
d9cf55bce9 LLM: fix MLP check of mixtral (#9891) 2024-01-11 18:01:59 +08:00
Ziteng Zhang
4af88a67b9 support chatglm3 with bf16 (#9888) 
* support chatglm3 with bigdl-bf16
 2024-01-11 16:45:21 +08:00
Yuwen Hu
0aef35a965 [LLM] Improve LLM doc regarding windows gpu related info (#9880) 
* Improve runtime configuration for windows

* Add python 310/311 supports for wheel downloading

* Add troubleshooting for windows gpu

* Remove manually import ipex due to auto importer

* Add info regarding cpu_embedding=True on iGPU

* More info for Windows users

* Small updates to API docs

* Python style fix

* Remove tip for loading from saved optimize_model for now

* Updated based on comments

* Update win info for multi-intel gpus selection

* Small fix

* Small fix
 2024-01-11 14:37:16 +08:00
Ruonan Wang
53531ae4ee LLM: support qkv fusion for fp8e5 (#9878) 
* update

* add mistral

* meet code review
 2024-01-10 17:50:00 +08:00
Lilac09
cb32b985ec add mistral and chatglm support to vllm (#9879) 
* add mistral and chatglm support to vllm

* add mistral and chatglm support to vllm
 2024-01-10 15:38:42 +08:00
Ruonan Wang
3e05c9e11b LLM: update esimd sdp kernel (#9871) 2024-01-09 18:10:01 +08:00
Yishuo Wang
36496d60ac only use quantize kv cache on MTL (#9862) 2024-01-09 13:24:02 +08:00
ZehuaCao
146076bdb5 Support llm-awq backend (#9856) 
* Support for LLM-AWQ Backend

* fix

* Update README.md

* Add awqconfig

* modify init

* update

* support llm-awq

* fix style

* fix style

* update

* fix AwqBackendPackingMethod not found error

* fix style

* update README

* fix style

---------

Co-authored-by: Uxito-Ada <414416158@qq.com>
Co-authored-by: Heyang Sun <60865256+Uxito-Ada@users.noreply.github.com>
Co-authored-by: cyita <yitastudy@gmail.com>
 2024-01-09 13:07:32 +08:00
Ruonan Wang
fea6f16057 LLM: add mlp fusion for fp8e5 and update related check (#9860) 
* update mlp fusion

* fix style

* update
 2024-01-09 09:56:32 +08:00
Jiao Wang
3b6372ab12 Fix Llama transformers 4.36 support (#9852) 
* supoort 4.36

* style

* update

* update

* update

* fix merge

* update
 2024-01-08 00:32:23 -08:00
Chen, Zhentao
1b585b0d40 set fp8 default as e5m2 (#9859) 2024-01-08 15:53:57 +08:00
Ruonan Wang
dc995006cc LLM: add flash attention for mistral / mixtral (#9846) 
* add flash attention for mistral

* update

* add flash attn for mixtral

* fix style
 2024-01-08 09:51:34 +08:00
Yishuo Wang
afaa871144 [LLM] support quantize kv cache to fp8 (#9812) 2024-01-08 09:28:20 +08:00
Jiao Wang
248ae7fad2 LLama optimize_model to support transformers 4.36 (#9818) 
* supoort 4.36

* style

* update

* update

* update
 2024-01-05 11:30:18 -08:00
Ruonan Wang
a60bda3324 LLM: update check for deepspeed (#9838) 2024-01-05 16:44:10 +08:00
Ruonan Wang
16433dd959 LLM: fix first token judgement of flash attention (#9841) 
* fix flash attention

* meet code review

* fix
 2024-01-05 13:49:37 +08:00
Yina Chen
f919f5792a fix kv cache out of bound (#9827) 2024-01-05 12:38:57 +08:00
Ruonan Wang
5df31db773 LLM: fix accuracy issue of chatglm3 (#9830) 
* add attn mask for first token

* fix

* fix

* change attn calculation

* fix

* fix

* fix style

* fix style
 2024-01-05 10:52:05 +08:00
Xiangyu Tian
38c05be1c0 [LLM] Fix dtype mismatch in Baichuan2-13b (#9834) 2024-01-04 15:34:42 +08:00
Ziteng Zhang
05b681fa85 [LLM] IPEX auto importer set on by default (#9832) 
* Set BIGDL_IMPORT_IPEX default to True

* Remove import intel_extension_for_pytorch as ipex from GPU example
 2024-01-04 13:33:29 +08:00
Wang, Jian4
4ceefc9b18 LLM: Support bitsandbytes config on qlora finetune (#9715) 
* test support bitsandbytesconfig

* update style

* update cpu example

* update example

* update readme

* update unit test

* use bfloat16

* update logic

* use int4

* set defalut bnb_4bit_use_double_quant

* update

* update example

* update model.py

* update

* support lora example
 2024-01-04 11:23:16 +08:00
Ruonan Wang
20e9742fa0 LLM: fix chatglm3 issue (#9820) 
* fix chatglm3 issue

* small update
 2024-01-03 16:15:55 +08:00
Wang, Jian4
a54cd767b1 LLM: Add gguf falcon (#9801) 
* init falcon

* update convert.py

* update style
 2024-01-03 14:49:02 +08:00
Qiyuan Gong
f0f9d45eac [LLM] IPEX import support bigdl-core-xe-21 (#9769) 
Add support for bigdl-core-xe-21.
 2023-12-28 15:23:58 +08:00
Guancheng Fu
5857a38321 [vLLM] Add option to adjust KV_CACHE_ALLOC_BLOCK_LENGTH (#9782) 
* add option kv_cache_block

* change var name
 2023-12-28 14:41:47 +08:00
Ruonan Wang
99bddd3ab4 LLM: better FP16 support for Intel GPUs (#9791) 
* initial support

* fix

* fix style

* fix

* limi esimd usage condition

* refactor code

* fix style

* small fix

* meet code review

* small fix
 2023-12-28 13:30:13 +08:00
Yishuo Wang
7d9f6c6efc fix cpuinfo error (#9793) 2023-12-28 09:23:44 +08:00
Wang, Jian4
7ed9538b9f LLM: support gguf mpt (#9773) 
* add gguf mpt

* update
 2023-12-28 09:22:39 +08:00
Cengguang Zhang
d299f108d0 update falcon attention forward. (#9796) 2023-12-28 09:11:59 +08:00
Kai Huang
689889482c Reduce max_cache_pos to reduce Baichuan2-13B memory (#9694) 
* optimize baichuan2 memory

* fix

* style

* fp16 mask

* disable fp16

* fix style

* empty cache

* revert empty cache
 2023-12-26 19:51:25 +08:00
Xiangyu Tian
0ea842231e [LLM] vLLM: Add api_server entrypoint (#9783) 
Add vllm.entrypoints.api_server for benchmark_serving.py in vllm.
 2023-12-26 16:03:57 +08:00
Ruonan Wang
11d883301b LLM: fix wrong batch output caused by flash attention (#9780) 
* fix

* meet code review

* move batch size check to the beginning

* move qlen check inside function

* meet code review
 2023-12-26 09:41:27 +08:00
Heyang Sun
66e286a73d Support for Mixtral AWQ (#9775) 
* Support for Mixtral AWQ

* Update README.md

* Update README.md

* Update awq_config.py

* Update README.md

* Update README.md
 2023-12-25 16:08:09 +08:00
Ruonan Wang
1917bbe626 LLM: fix BF16Linear related training & inference issue (#9755) 
* fix bf16 related issue

* fix

* update based on comment & add arc lora script

* update readme

* update based on comment

* update based on comment

* update

* force to bf16

* fix style

* move check input dtype into function

* update convert

* meet code review

* meet code review

* update merged model to support new training_mode api

* fix typo
 2023-12-25 14:49:30 +08:00
Xiangyu Tian
30dab36f76 [LLM] vLLM: Fix kv cache init (#9771) 
Fix kv cache init
 2023-12-25 14:17:06 +08:00
Yina Chen
449b387125 Support relora in bigdl-llm (#9687) 
* init

* fix style

* update

* support resume & update readme

* update

* update

* remove important

* add training mode

* meet comments
 2023-12-25 14:04:28 +08:00
Ziteng Zhang
986f65cea9 [LLM] Add trust_remote_code for local renamed model in bigdl_llm_model.py (#9762) 2023-12-25 11:31:14 +08:00
Guancheng Fu
daf536fb2d vLLM: Apply attention optimizations for selective batching (#9758) 
* fuse_rope for prefil

* apply kv_cache optimizations

* apply fast_decoding_path

* Re-enable kv_cache optimizations for prefill

* reduce KV_CACHE_ALLOC_BLOCK for selective_batching
 2023-12-25 10:29:31 +08:00
Qiyuan Gong
4c487313f2 Revert "[LLM] IPEX auto importer turn on by default for XPU (#9730)" (#9759) 
This reverts commit 0284801fbd.
 2023-12-22 16:38:24 +08:00
Qiyuan Gong
0284801fbd [LLM] IPEX auto importer turn on by default for XPU (#9730) 
* Set BIGDL_IMPORT_IPEX default to true, i.e., auto import IPEX for XPU.
* Remove import intel_extension_for_pytorch as ipex from GPU example.
* Add support for bigdl-core-xe-21.
 2023-12-22 16:20:32 +08:00
Guancheng Fu
fdf93c9267 Implement selective batching for vLLM (#9659) 
* add control to load hf model

* finish initial version of selective_batching

* temp

* finish

* Remove print statement

* fix error

* Apply yang's optimization

* a version that works

* We need to check kv_cache passed in, this could be an error. TODO: add fast decoding path

* format

* temp solution: not batching prefill requests

* a version that works for prefill batching

* format

* a solid version: works normally

* a temp version

* Solid version: remove redundant functions

* fix format

* format

* solid: add option to enable selective_batching

* remove logic for using transformer models

* format

* format

* solid: enable argument VLLM_ENABLE_SELECTIVE_BATCHING

* format

* finish

* format
 2023-12-22 13:45:46 +08:00
Ruonan Wang
2f36769208 LLM: bigdl-llm lora support & lora example (#9740) 
* lora support and single card example

* support multi-card, refactor code

* fix model id and style

* remove torch patch, add two new class for bf16, update example

* fix style

* change to training_mode

* small fix

* add more info in help

* fixstyle, update readme

* fix ut

* fix ut

* Handling compatibility issues with default LoraConfig
 2023-12-22 11:05:39 +08:00
SONG Ge
ba0b939579 [LLM] Support transformers-v4.36.0 on mistral model (#9744) 
* add support transformers-v4.36.0 on mistral model

* python/llm/src/bigdl/llm/transformers/models/mistral.py

* make the redundant implementation as utils

* fix code style

* fix

* fix style

* update with utils enough_kv_room
 2023-12-22 09:59:27 +08:00
Xin Qiu
e36111e713 mixstral fused qkv and rope (#9724) 
* mixstral fused qkv and rope

* fix and clean

* fix style

* update

* update

* fix

* update

* fix
 2023-12-22 09:26:35 +08:00
Jiao Wang
e4f6e43675 safetenor to false (#9728) 2023-12-21 14:41:51 -08:00
Yishuo Wang
426660b88e simplify qwen attention (#9747) 2023-12-21 17:53:29 +08:00
Wang, Jian4
984697afe2 LLM: Add bloom gguf support (#9734) 
* init

* update bloom add merges

* update

* update readme

* update for llama error

* update
 2023-12-21 14:06:25 +08:00
Heyang Sun
df775cf316 fix python style (#9742) 
* fix python style

* fix

* fix
 2023-12-21 11:25:05 +08:00
Xin Qiu
6c3e698bf1 mistral decoding_fast_path and fused mlp (#9714) 
* mistral decoding_fast_path and fused mlp

* meet code review
 2023-12-21 10:11:37 +08:00
Heyang Sun
d157f623b6 Load Mixtral gguf in a block-wise way (#9725) 
* Load Mixtral gguf in a block-wise way

* refine
 2023-12-21 10:03:23 +08:00
Zhao Changmin
4bda975a3e LLM: Align lowbit model config (#9735) 
* align lowbit model config
 2023-12-21 09:48:58 +08:00
Wang, Jian4
e1e921f425 LLM: gguf other model using dtype (#9729) 2023-12-21 09:33:40 +08:00
Yishuo Wang
13ea6330bd optimize qwen rope (#9737) 2023-12-20 17:34:34 +08:00
Ziteng Zhang
4c032a433e [LLM] Add glibc checker (#9624) 
* Add glibc checker
* Add env BIGDL_GLIBC_CHECK to control glibc checker. The default is false, i.e., don't check.
 2023-12-20 16:52:43 +08:00
Yina Chen
cd652a1710 Support fp8 e5m2 on arc (#9711) 
* init

* fix style

* update

* fix style

* update
 2023-12-20 16:26:17 +08:00
Yishuo Wang
e54c428d30 add bf16/fp16 fuse mlp support (#9726) 2023-12-20 10:40:45 +08:00
Heyang Sun
612651cb5d fix typo (#9723) 2023-12-20 09:41:59 +08:00
Yishuo Wang
522cf5ed82 [LLM] Improve chatglm2/3 rest token performance with long context (#9716) 2023-12-19 17:29:38 +08:00
Yishuo Wang
f2e6abb563 fix mlp batch size check (#9718) 2023-12-19 14:22:22 +08:00
Heyang Sun
1fa7793fc0 Load Mixtral GGUF Model (#9690) 
* Load Mixtral GGUF Model

* refactor

* fix empty tensor when to cpu

* update gpu and cpu readmes

* add dtype when set tensor into module
 2023-12-19 13:54:38 +08:00
Qiyuan Gong
d0a3095b97 [LLM] IPEX auto importer (#9706) 
* IPEX auto importer and get_ipex_version.
* Add BIGDL_IMPORT_IPEX to control auto import, default is false.
 2023-12-19 13:39:38 +08:00
Yang Wang
f4fb58d99c fusing qkv project and rope (#9612) 
* Try fusing qkv project and rope

* add fused mlp

* fuse append cache

* fix style and clean up code

* clean up
 2023-12-18 16:45:00 -08:00
Cengguang Zhang
4d22add4af LLM: fix qwen efficiency issue in perf-test. 2023-12-18 18:32:54 +08:00
Ruonan Wang
8ed89557e5 LLM: add mlp optimization of mixtral (#9709) 2023-12-18 16:59:52 +08:00
Xin Qiu
320110d158 handle empty fused norm result (#9688) 
* handle empty fused norm result

* remove fast_rms_norm

* fix style
 2023-12-18 09:56:11 +08:00
SONG Ge
d5b81af7bd Support mixtral attention optimization on transformers-v4.36.0 (#9674) 
* add example code to support mistral/mixtral attention on transformers v4.36.0

* update

* style fix

* add update for seen-tokens

* support mixtral

* rm mistral change

* small fix

* add more comments and remove use_cache part

---------

Co-authored-by: plusbang <binbin1.deng@intel.com>
 2023-12-15 14:30:23 +08:00
Cengguang Zhang
adbef56001 LLM: update qwen attention forward. (#9695) 
* feat: update qwen attention forward.

* fix: style.
 2023-12-15 14:06:15 +08:00
Wang, Jian4
b8437a1c1e LLM: Add gguf mistral model support (#9691) 
* add mistral support

* need to upgrade transformers version

* update
 2023-12-15 13:37:39 +08:00
Wang, Jian4
496bb2e845 LLM: Support load BaiChuan model family gguf model (#9685) 
* support baichuan model family gguf model

* update gguf generate.py

* add verify models

* add support model_family

* update

* update style

* update type

* update readme

* update

* remove support model_family
 2023-12-15 13:34:33 +08:00
Yishuo Wang
9a330bfc2b fix fuse mlp when using q5_0 or fp8 (#9689) 2023-12-14 16:16:05 +08:00
Xin Qiu
5e46e0e5af fix baichuan2-7b 1st token performance regression on xpu (#9683) 
* fix baichuan2-7b 1st token performance regression

* add comments

* fix style
 2023-12-14 09:58:32 +08:00
Yishuo Wang
09ca540f9b use fuse mlp in qwen (#9672) 2023-12-13 17:20:08 +08:00
Ruonan Wang
c7741c4e84 LLM: update moe block convert to optimize rest token latency of Mixtral (#9669) 
* update moe block convert

* further accelerate final_hidden_states

* fix style

* fix style
 2023-12-13 16:17:06 +08:00
Xiangyu Tian
1c6499e880 [LLM] vLLM: Support Mixtral Model (#9670) 
Add Mixtral support for BigDL vLLM.
 2023-12-13 14:44:47 +08:00
Ruonan Wang
dc5b1d7e9d LLM: integrate sdp kernel for FP16 rest token inference on GPU [DG2/ATSM] (#9633) 
* integrate sdp

* update api

* fix style

* meet code review

* fix

* distinguish mtl from arc

* small fix
 2023-12-13 11:29:57 +08:00
Qiyuan Gong
5b0e7e308c [LLM] Add support for empty activation (#9664) 
* Add support for empty activation, e.g., [0, 4096]. Empty activation is allowed by PyTorch.
* Add comments.
 2023-12-13 11:07:45 +08:00
SONG Ge
284e7697b1 [LLM] Optimize ChatGLM2 kv_cache to support beam_search on ARC (#9579) 
* optimize kv_cache to support beam_search on Arc

* correctness test update

* fix query_length issue

* simplify implementation

* only enable the optimization on gpu device

* limit the beam_search support only enabled with gpu device and batch_size > 1

* add comments for beam_search case and revert ut change

* meet comments

* add more comments to describe the differece between multi-cases
 2023-12-13 11:02:14 +08:00
Ziteng Zhang
8931f2eb62 [LLM] Fix transformer qwen size mismatch and rename causal_mask (#9655) 
* Fix size mismatching caused by context_layer
* Change registered_causal_mask to causal_mask
 2023-12-12 20:57:40 +08:00
binbin Deng
59ce86d292 LLM: support optimize_model=True for Mixtral model (#9657) 2023-12-12 16:41:26 +08:00
Heyang Sun
9f02f96160 [LLM] support for Yi AWQ model (#9648) 2023-12-11 14:07:34 +08:00
Xin Qiu
82255f9726 Enable fused layernorm (#9614) 
* bloom layernorm

* fix

* layernorm

* fix

* fix

* fix

* style fix

* fix

* replace nn.LayerNorm
 2023-12-11 09:26:13 +08:00
Yina Chen
70f5e7bf0d Support peft LoraConfig (#9636) 
* support peft loraconfig

* use testcase to test

* fix style

* meet comments
 2023-12-08 16:13:03 +08:00
Xin Qiu
0b6f29a7fc add fused rms norm for Yi and Qwen (#9640) 2023-12-08 16:04:38 +08:00
Xin Qiu
5636b0ba80 set new linear status (#9639) 2023-12-08 11:02:49 +08:00
Yuwen Hu
6f34978b94 [LLM] Add more performance tests for win iGPU (more in-out pairs, RWKV model) (#9626) 
* Add supports for loading rwkv models using from_pretrained api

* Temporarily enable pr tests

* Add RWKV in tests and more in-out pairs

* Add rwkv for 512 tests

* Make iterations smaller

* Change back to nightly trigger
 2023-12-07 18:55:16 +08:00
Ruonan Wang
d9b0c01de3 LLM: fix unlora module in qlora finetune (#9621) 
* fix unlora module

* split train and inference
 2023-12-07 16:32:02 +08:00
Yishuo Wang
7319f2c227 use fused mlp in baichuan2 (#9620) 2023-12-07 15:50:57 +08:00
Xiangyu Tian
deee65785c [LLM] vLLM: Delete last_kv_cache before prefilling (#9619) 
Remove last_kv_cache before prefilling to reduce peak memory usage.
 2023-12-07 11:32:33 +08:00
Xiangyu Tian
0327169b50 [LLM] vLLM: fix memory leak in prepare_kv_cache (#9616) 
Revert modification in prepare_kv_cache to fix memory leak.
 2023-12-07 10:08:18 +08:00
Xin Qiu
13d47955a8 use fused rms norm in chatglm2 and baichuan (#9613) 
* use fused rms norm in chatglm2 and baichuan

* style fix
 2023-12-07 09:21:41 +08:00
Yina Chen
404e101ded QALora example (#9551) 
* Support qa-lora

* init

* update

* update

* update

* update

* update

* update merge

* update

* fix style & update scripts

* update

* address comments

* fix typo

* fix typo

---------

Co-authored-by: Yang Wang <yang3.wang@intel.com>
 2023-12-06 15:36:21 +08:00
Guancheng Fu
6978b2c316 [VLLM] Change padding patterns for vLLM & clean code (#9609) 
* optimize

* fix minor error

* optimizations

* fix style
 2023-12-06 15:27:26 +08:00
Zheng, Yi
d154b38bf9 Add llama2 gpu low memory example (#9514) 
* Add low memory example

* Minor fixes

* Update readme.md
 2023-12-05 17:29:48 +08:00
Ziteng Zhang
65934c9f4f [LLM] Fix Qwen causal_mask and attention_mask size mismatching (#9600) 
* Fix #9582 , caused by Qwen modified modeling_qwen.py 7f62181c94 (d2h-049182)
 2023-12-05 15:15:54 +08:00
Qiyuan Gong
f211f136b6 Configurable TORCH_LINEAR_THRESHOLD from env (#9588) 
* Add TORCH_LINEAR_THRESHOLD from env (BIGDL_LLM_LINEAR_THRESHOLD)
* Change default to 512
 2023-12-05 13:19:47 +08:00
Xiangyu Tian
5c03651309 [LLM] vLLM: Add Preempt for scheduler (#9568) 
Implement Preempt_by_recompute method for vllm.
 2023-12-03 20:16:25 +08:00
Xin Qiu
69c49d21f5 use fused rms norm (#9572) 
* use fused rms norm

* meet code review
 2023-11-30 21:47:41 +08:00
Yishuo Wang
7f6465518a support loading llama tokenizer from gguf model (#9565) 2023-11-30 14:56:12 +08:00
Yuwen Hu
34503efa6a Fix cpu pinned embedding (#9556) 2023-11-29 18:27:56 +08:00
binbin Deng
4ff2ca9d0d LLM: fix loss error on Arc (#9550) 2023-11-29 15:16:18 +08:00
Yishuo Wang
65121c7997 support loading q4_1/q5_0/q5_1/q8_0 gguf model (#9546) 2023-11-29 14:40:37 +08:00
Yuwen Hu
5f5ca38b74 [LLM Doc] Fix api doc rendering error (#9542) 
* Fix api rendering error

* Fix python style
 2023-11-29 09:17:09 +08:00
Yishuo Wang
a86c6e0b56 [LLM] support loading gguf model (#9544) 2023-11-28 15:51:15 +08:00
Xiangyu Tian
916c338772 fix bugs in vllm length check (#9543) 2023-11-28 11:09:54 +08:00
Zhao Changmin
e7e0cd3b5e CPU Pinned embedding Layer (#9538) 
* CPU Pinned embedding
 2023-11-28 09:46:31 +08:00
Guancheng Fu
963a5c8d79 Add vLLM-XPU version's README/examples (#9536) 
* test

* test

* fix last kv cache

* add xpu readme

* remove numactl for xpu example

* fix link error

* update max_num_batched_tokens logic

* add explaination

* add xpu environement version requirement

* refine gpu memory

* fix

* fix style
 2023-11-28 09:44:03 +08:00
Guancheng Fu
b6c3520748 Remove xformers from vLLM-CPU (#9535) 2023-11-27 11:21:25 +08:00
binbin Deng
6bec0faea5 LLM: support Mistral AWQ models (#9520) 2023-11-24 16:20:22 +08:00
Ruonan Wang
914a5a5a27 LLM: fix abnormal Mistral GPU accuracy by updating rms_norm (#9529) 2023-11-24 15:37:50 +08:00
SONG Ge
3d24823cda hot-fix mistral kv_cache (#9528) 2023-11-24 14:33:04 +08:00
Zhao Changmin
42b7a16bc5 Replace torch.bmm with safe_bmm (#9519) 
* replace bmm with safe one

* rename args and deprecated warning
 2023-11-24 12:16:48 +08:00
Ruonan Wang
b63aae8a8e LLM: add flash attention support for llama (#9518) 
* add initial flash attention for llama

* accelerate fp32 first token by changing to fp16 in advance

* support fp32
 2023-11-23 18:40:18 +08:00
Guancheng Fu
bf579507c2 Integrate vllm (#9310) 
* done

* Rename structure

* add models

* Add structure/sampling_params,sequence

* add input_metadata

* add outputs

* Add policy,logger

* add and update

* add parallelconfig back

* core/scheduler.py

* Add llm_engine.py

* Add async_llm_engine.py

* Add tested entrypoint

* fix minor error

* Fix everything

* fix kv cache view

* fix

* fix

* fix

* format&refine

* remove logger from repo

* try to add token latency

* remove logger

* Refine config.py

* finish worker.py

* delete utils.py

* add license

* refine

* refine sequence.py

* remove sampling_params.py

* finish

* add license

* format

* add license

* refine

* refine

* Refine line too long

* remove exception

* so dumb style-check

* refine

* refine

* refine

* refine

* refine

* refine

* add README

* refine README

* add warning instead error

* fix padding

* add license

* format

* format

* format fix

* Refine vllm dependency (#1)

vllm dependency clear

* fix licence

* fix format

* fix format

* fix

* adapt LLM engine

* fix

* add license

* fix format

* fix

* Moving README.md to the correct position

* Fix readme.md

* done

* guide for adding models

* fix

* Fix README.md

* Add new model readme

* remove ray-logic

* refactor arg_utils.py

* remove distributed_init_method logic

* refactor entrypoints

* refactor input_metadata

* refactor model_loader

* refactor utils.py

* refactor models

* fix api server

* remove vllm.stucture

* revert by txy 1120

* remove utils

* format

* fix license

* add bigdl model

* Refer to a specfic commit

* Change code base

* add comments

* add async_llm_engine comment

* refine

* formatted

* add worker comments

* add comments

* add comments

* fix style

* add changes

---------

Co-authored-by: xiangyuT <xiangyu.tian@intel.com>
Co-authored-by: Xiangyu Tian <109123695+xiangyuT@users.noreply.github.com>
Co-authored-by: leonardozcm <leonardo1997zcm@gmail.com>
 2023-11-23 16:46:45 +08:00
Qiyuan Gong
0f0c6bb631 [LLM] Fix Qwen registered_causal_mask is None (#9513) 
* Add registered_causal_mask init based on 2abd8e5777.
 2023-11-23 09:28:04 +08:00
Ruonan Wang
076d106ef5 LLM: GPU QLoRA update to bf16 to accelerate gradient checkpointing (#9499) 
* update to bf16 to accelerate gradient checkpoint

* add utils and fix ut
 2023-11-21 17:08:36 +08:00
Xin Qiu
50b01058f1 enable new q4_1 (#9479) 2023-11-17 14:58:57 +08:00
Zhao Changmin
30abd304a7 LLM: Fix baichuan pre-normalize model tensor assigning issue when loading (#9481) 
* No need to normalized when loading
 2023-11-16 21:57:28 +08:00
Ruonan Wang
c0ef70df02 llm: quick fix of fast_rms_norm (#9480) 2023-11-16 14:42:16 +08:00